Make a node-level dataset from a block-level dataset

Given the results of the splitting and testing algorithm, make a node level data set for use in reporting results and as input to ggraph for visualization in terms of a tree graph.

Usage

make_results_tree(
  orig_res,
  block_id,
  node_label = NULL,
  return_what = "all",
  truevar_name = NULL,
  node_dat = NULL,
  node_tracker = NULL
)

Arguments

orig_res

data.table from find_blocks(); must include elements such as

• node_id (node identifier, may be factorized),

• p1,p2,… and a pfinal*,

• alpha1, alpha2, …

block_id

name of your block ID column (e.g. "bF")

node_label

optional name of a descriptive label column

return_what

a character vector containing "all", "graph" (a tbl_graph object with nodes and edges), "nodes" (a data.table with node level information), "test_summary" (a data.table object with one row indicating false and true discoveries, etc.)

truevar_name

the optional name of a column recording the true treatment effect (used here to find blocks where the true effect is 0 or not). In some simulations we have a column called nonnull which is TRUE if that block or node has a non-zero effect and FALSE if the block or node has a truly zero effect. So, truevar_name can be "nonnull"

node_dat

optional node-level output from find_blocks(). When the full list returned by find_blocks() is supplied as orig_res this argument is populated automatically. When providing only the block-level data.table you must also pass node_dat so that node identifiers can be matched without relying on p-values.

node_tracker

optional tracker returned by find_blocks(). This is not required for constructing the node table but is accepted for compatibility.

Value

a list that can contain nodes, a tbl_graph object, and/or a test_summary

Examples

if (FALSE) { # \dontrun{
# Complete workflow example
data(example_dat, package = "manytestsr")
library(data.table)
library(dplyr)

# Create block-level dataset
example_bdat <- example_dat %>%
  group_by(blockF) %>%
  summarize(
    nb = n(),
    pb = mean(trt),
    hwt = (nb / nrow(example_dat)) * (pb * (1 - pb)),
    .groups = "drop"
  ) %>%
  as.data.table()

# Run find_blocks
results <- find_blocks(
  idat = example_dat,
  bdat = example_bdat,
  blockid = "blockF",
  splitfn = splitCluster,
  pfn = pOneway,
  fmla = Y1 ~ trtF | blockF,
  parallel = "no"
)

# Create tree structure (default returns all components)
tree_results <- make_results_tree(results, block_id = "blockF")

# Examine the components
str(tree_results)

# Look at node-level information
head(tree_results$nodes)

# Look at test summary
print(tree_results$test_summary)

# Get only the graph component
tree_graph <- make_results_tree(results,
  block_id = "blockF",
  return_what = "graph"
)
print(tree_graph)

# Get only node information
tree_nodes <- make_results_tree(results,
  block_id = "blockF",
  return_what = "nodes"
)
head(tree_nodes)
} # }